Re: [p2prg] Fwd: I-D Action:draft-irtf-p2prg-alto-survey-01.txt

[Enrico Marocco <[e-mail]>]

Ok, we'll integrate the text in an update that will be out soon.

Enrico

Arnaud Legout wrote:
> Hi,
> 
> I am fine with the text you propose, except that I would remove the sentence
> "especially in networks with lower density of peers." I believe that it 
> is actually the opposite.
> When there is a low density of peers, the locality policy will have few 
> impact on the overlay
> interconnecting peers.
> Find below a revised text.
> 
> You are right that my points better fit into the mythbusting draft.
> I will send you a proposition of text in a separate email.
> 
> Thanks,
> Arnaud.
> 
>    The literature presented in Section 2 shows that a certain level of
>    locality-awareness in the peer selection process of P2P algorithms is
>    usually beneficial to the application performance.  However, an 
>    excessive localization of the traffic might cause partitioning in 
>    the overlay interconnecting peers, which will negatively affect 
>    the performance experienced by the peers.
> 
>    Finding the right balance between localization and randomness in peer
>    selection is an open issue. At the time of writing, it seems that
>    different applications have different levels of tolerance and should
>    be addressed separately.  Le Blond et al. [LeBlond] have studied the
>    specific case of BitTorrent, proposing a simple mechanism to prevent
>    partitioning in the overlay, yet reaching a high level of cross-domain
>    traffic reduction without adversely impacting peers.
> 
> 
> 
> Enrico Marocco a écrit :
>> Hi Arnoud,
>>
>> the work you mention is actually relevant to ALTO, but I'm not sure how
>> to fit it in the survey part of the draft (that, in fact, is about
>> proposed solutions). However, it deals with a relevant issue we did not
>> mention, so what about adding the following text in a subsection of
>> section 4?
>>
>>
>> 4.6.  Negative impact of over-localization
>>
>>    The literature presented in Section 2 show that certain level of
>>    locality-awareness in the peer selection process of P2P algorithms is
>>    usually beneficial to the application performance.  However, it seems
>>    reasonable to expect that an excessive impact of localization on such
>>    a process (to the detriment of randomness) will cause partitioning in
>>    overlay networks that will negatively affect the performance
>>    experienced by the users, especially in networks with lower density
>>    of peers.
>>
>>    Finding the right balance between localization and randomness in peer
>>    selection is an open issue and, at the time of writing, it seems that
>>    different applications have different levels of tolerance and should
>>    be addressed separately.  Le Blond et al. [LeBlond] have studied the
>>    specific case of BitTorrent, proposing a simple mechanism to prevent
>>    partitioning in the overlay, yet reaching high levels of cross-domain
>>    traffic reduction without adversely impacting peers.
>>
>>
>> I'd rather avoid the details of how much traffic could be reduced
>> according to your experience, as that's not really the goal of this
>> document (it is actually mythbusting draft's goal). WDYT?
>>
>> Enrico
>>
>> Arnaud Legout wrote:
>>   
>>> Hi,
>>>
>>> I believe that our work on "Pushing BitTorrent Locality to the Limit" 
>>> http://hal.inria.fr/inria-00343822/en/
>>> is highly relevant to ALTO.
>>> Whereas we do not propose a new architecture, we perform a thorough 
>>> large scale evaluation using real world
>>> data of the impact of locality. In particular, we answer two important 
>>> questions for ALTO.
>>>
>>> First, how much traffic can be kept local without adversely impacting 
>>> peers? Whatever the ALTO architecture is,
>>> it will impact the structure of the overlay interconnecting peers.
>>> We go much further than previous work on the understanding of the impact 
>>> of locality on the structure of that overlay.
>>> In addition, reducing the amount of traffic that is kept local in order 
>>> to prevent partitions (thus a loss of performance for peers)
>>> is the solution adopted by P4P and Ono. We introduce a simple mechanism 
>>> (that is backward compatible) to prevent partitions
>>> and show that the traffic reduction on specific links can be 
>>> dramatically reduced by keeping more traffic local without adversely
>>> impacting peers. Those kind of mechanisms are very important because 
>>> they enable to reap full benefits from the information
>>> provided by ALTO.
>>>
>>> Second, what would be the benefit of a locality policy at the Internet 
>>> scale?
>>> We show using a real world crawl of a large fraction of the all 
>>> BitTorrent peers that the benefit from deploying
>>> a locality policy today would be a reduction of 40% on inter-AS links. 
>>> To the best of our knowledge, the work closer to ours is
>>> the one of P4P. However, they only consider one torrent and one AS. 
>>> Whereas the P4P field tests are successfully used to support the relevance
>>> of the P4P architecture, they cannot be used to support the relevance of 
>>> a locality policy at the scale of the Internet. But that last
>>> point fundamental for the ALTO justification. A classical
>>> argument (that we often faced before performing the large scale 
>>> BitTorrent measurement) is that at the scale of the Internet
>>> the benefit of a locality policy will be negligible because there is on 
>>> average one or a few peers per AS. We show that even if
>>> it is true that there is on average few peers per AS, the reduction of 
>>> traffic on inter-AS links that can be achieved using a locality
>>> policy is still high at the Internet scale. Thus we believe that this is 
>>> the first large scale measurement that strongly support the relevance of the
>>> ALTO effort.
>>>
>>> Finally, we do not claim that inter-AS links are always the right links 
>>> on which to perform traffic reduction, but that it is a reasonable first
>>> level approximation.
>>>
>>> I hope those comments will help.
>>>
>>> Arnaud.
>>>
>>> stefano previdi a écrit :
>>>     
>>>> Folks,
>>>>
>>>> the authors have submitted an updated version of alto-survey draft.
>>>> Please, bring to the list your comments and we'll proceed with the
>>>> publication process.
>>>>
>>>> Thanks.
>>>>
>>>> s.
>>>>
>>>>
>>>> Begin forwarded message:
>>>>
>>>>       
>>>>> From: [e-mail]
>>>>> Date: December 11, 2009 4:30:02 PM GMT+01:00
>>>>> To: [e-mail]
>>>>> Subject: I-D Action:draft-irtf-p2prg-alto-survey-01.txt
>>>>> Reply-To: [e-mail]
>>>>>
>>>>> A New Internet-Draft is available from the on-line Internet-Drafts 
>>>>> directories.
>>>>>
>>>>>     Title           : A Survey on Research on the Application-Layer 
>>>>> Traffic Optimization (ALTO) Problem
>>>>>     Author(s)       : I. Rimac, et al.
>>>>>     Filename        : draft-irtf-p2prg-alto-survey-01.txt
>>>>>     Pages           : 17
>>>>>     Date            : 2009-12-11
>>>>>
>>>>> A significant part of the Internet traffic today is generated by
>>>>> peer-to-peer (P2P) applications used traditionally for file-sharing,
>>>>> and more recently for real-time communications and live media
>>>>> streaming.  Such applications discover a route to each other through
>>>>> an overlay network with little knowledge of the underlying network
>>>>> topology.  As a result, they may choose peers based on information
>>>>> deduced from empirical measurements, which can lead to suboptimal
>>>>> choices.  This document, a product of the P2P Research Group,
>>>>> presents a survey of existing literature on discovering and using
>>>>> network topology information for application-layer traffic
>>>>> optimization.
>>>>>         
>>> _______________________________________________
>>> p2prg mailing list
>>> [e-mail]
>>> http://www.irtf.org/mailman/listinfo/p2prg